SerialLain3170/adeleine

What it solves

Adeleine is a deep-learning framework for automatically colorizing line art. It addresses the challenge of making colorization consistent and flexible, allowing users to guide the process using various types of hints to achieve a specific desired look.

How it works

The project implements several image-to-image translation methods (such as pix2pix, pix2pixHD, and BicycleGAN) and supports three primary ways to provide guidance (hints) for colorization:

• Atari hints: Using specific colored lines to indicate desired colors.
• Tag hints: Using text tags to guide the color palette.
• Reference hints: Using existing images as a style or color reference.

To prevent the model from overfitting to a specific style of line art, it utilizes multiple line extraction methods (XDoG, SketchKeras, and Sketch Simplification) and applies data augmentation techniques like random morphology transformations and RGB value adjustments.

Who it’s for

Digital artists and illustrators who want to automate the process of filling colors into line art sketches, whether they are pencil drawings or digital drawings.

Highlights

• Flexible Guidance: Supports colorization via colored hints, text tags, or reference images.
• Robust Line Art Handling: Uses multiple extraction methods to ensure the model works across different drawing styles.
• Diverse Methodologies: Implements a variety of GAN-based architectures for both hint-less and guided colorization.
• GUI Application: Includes a graphical user interface for easier access to the tools.